Sanitize JobInternal

[lpellegr]

InternalJob implementation is really weird.

Several changes should be made to improve readability and performance:

• Replace HashMap implementation for tracking tasks by a TreeMap that orders entries by keys. Order seems already maintained by returning task id as hashcode. However, tasks are sometimes reordered and reinjected in the map (e.g. InternalJob#prepareTasks). Besides, it makes some assumptions about how HashMap is implemented (open addressing vs separate chaining)
• Check if prepareTasks method is still required once TreeMap is used
• Find where recreation and reinjection of task ids is made (e.g. in SchedulerDBManager#newJobSubmitted)
  • Check if these pieces of code are still required
• Check if it is possible to merge all different accessors for the internal hash map (e.g. getHMTasks, getITasks, getIHMTasks, getTasks, etc.). Ideally, a shallow copy should always be returned. If not possible, at least a better naming should be used.
• Check that accessors like getHMTasks (which returns a shallow copy) is not called inside a for loop multiple times when it is not required (like it was the case in SchedulerDBManager#newJobSubmitted).

[marcocast]

Instead of TreeMap, we may consider as well HashTreePMap from pCollection lib as it provides immutability and better performance together with clean code. http://pcollections.googlecode.com/svn/trunk/docs/org/pcollections/HashTreePMap.html

[lpellegr]

Interesting. ImmutableSortedMap from Guava should be also considered if it is possible to use an immutable map. This would avoid importing yet another library.

[fviale]

If ImmutableSortedMap refers to an immutable map, it would not be well suited to the tasks maps as tasks are added dynamically during iterations/replications. Unless I missed something.

[lpellegr]

@fviale yes, that's why I said if it is possible to use an immutable map. It was a suggestion related to what Marco proposed.

[fviale]

We should as well consider LinkedHashMap, which preserves insertion order. As TaskIds are always incrementing, insertion order should match TaskIds order.